Skip to main content
Be specific. Include details (what you want, where in the project, which scripts or objects) so the model can understand and act accurately. Use context. Add relevant info in the context panel or reference scripts by name so the agent can find and use them. Keep requests focused. One clear request usually works better than many at once. Break big goals into steps. Work in steps. Tackle one feature or system at a time instead of asking for “the whole game” in one go. Stay on one system per request. Focus on one area (e.g. movement, UI, saving) per turn; jumping between many systems in a single request is harder for the agent. Pick the right model. Newer models tend to be more thorough; choose based on your current task and quality vs. speed/cost. Choose the right Agent mode. Use Ask vs. Agent (and Debug when available) depending on whether you need answers only or actual changes in the project. Choose the right processing mode.
  • Fast: Uses the least context. Best for quick, narrow requests that don’t need full scene or hierarchy. Lowest token cost.
  • Default: Uses moderate context and does not include the full scene hierarchy. Good for general work; less suitable for debugging.
  • Thinking: Uses full available context. Best for debugging and complex tasks; higher token use per message.